In silico evo-devo: reconstructing stages in the evolution of animal segmentation
© The Author(s) 2016
Received: 21 April 2016
Accepted: 13 July 2016
Published: 1 August 2016
The evolution of animal segmentation is a major research focus within the field of evolutionary–developmental biology. Most studied segmented animals generate their segments in a repetitive, anterior-to-posterior fashion coordinated with the extension of the body axis from a posterior growth zone. In the current study we ask which selection pressures and ordering of evolutionary events may have contributed to the evolution of this specific segmentation mode.
To answer this question we extend a previous in silico simulation model of the evolution of segmentation by allowing the tissue growth pattern to freely evolve. We then determine the likelihood of evolving oscillatory sequential segmentation combined with posterior growth under various conditions, such as the presence or absence of a posterior morphogen gradient or selection for determinate growth. We find that posterior growth with sequential segmentation is the predominant outcome of our simulations only if a posterior morphogen gradient is assumed to have already evolved and selection for determinate growth occurs secondarily. Otherwise, an alternative segmentation mechanism dominates, in which divisions occur in large bursts through the entire tissue and all segments are created simultaneously.
Our study suggests that the ancestry of a posterior signalling centre has played an important role in the evolution of sequential segmentation. In addition, it suggests that determinate growth evolved secondarily, after the evolution of posterior growth. More generally, we demonstrate the potential of evo-devo simulation models that allow us to vary conditions as well as the onset of selection pressures to infer a likely order of evolutionary innovations.
KeywordsSegmentation In silico evolution Axis extension Posterior signalling Determinate growth Bilaterian evolution
Segmentation, the division of the animal body plan into multiple, repeating units, has fascinated evolutionary and developmental biologists alike. Only the vertebrates, arthropods and annelids display overt body segmentation, while several other clades show intermediate levels of segmentation in only a subset of tissues or organs, a property called metamerism [1–3]. Repetitive patterning is studied in most detail in overtly segmented animals. In these clades, segments are typically laid down in a regular anterior–posterior sequence, via a process involving posterior growth (also called terminal addition) and periodic, sequential generation of segments [3, 4]. A famous exception is the fruit fly Drosophila in which segments are laid down simultaneously across a preformed body axis.
It is currently unresolved why segmented animals mostly display this superficially similar, sequential mode of segmentation. This issue is partly related to the question of whether segmentation was present in the bilaterian ancestor, either as overt segmentation or as metamerism, or rather that it evolved multiple times in parallel in the different lineages [1, 2, 4–13]. Arguments in favour of a single origin of segmentation include the prevalence of sequential segmentation [3, 4]. Studies using ancestral state reconstruction suggest that this mode of segment addition via posterior outgrowth represents ancestral bilaterian properties [9, 14]. In addition, the three segmented lineages have shared genes involved in segmentation, such as Notch, Engrailed and Wnt [4, 8, 15, 16]. Arguments in favour of parallel evolution of segmentation instead put forward that there are also large differences in the genes responsible for segmentation and that the limited observed overlap in gene usage could be explained by parallel recruitment from the limited developmental genetic toolbox . The precise mechanisms of cell division, axial elongation and sequential segmentation also differ substantially between these lineages, ranging from teloblastic growth and stereotyped cell divisions in annelids and some crustaceans [17, 18], to posterior growth zones in most insects and vertebrates [19, 20] with variable roles of cell division versus cell rearrangement [21, 22]. Furthermore, multiple segmentation processes can take place in different body regions or tissue types even within a single organism, each with their own evolutionary origin , which further supports (partial) parallel evolution.
Still, independent of whether sequential segmentation evolved once or multiple times, we can ask whether certain factors or conditions may have contributed to this particular evolutionary outcome. Earlier evo-devo simulation studies have demonstrated that sequential segmentation represents a robust evolutionary outcome with high future evolutionary potential [24–26], suggesting evolutionary advantages of this particular segmentation mode. In addition, prior evolutionary events may have generated biases or constraints that influenced the likelihood of the evolution of sequential segmentation. For example, evolutionary comparisons show that a posterior signalling region characterised by caudal, Wnt and FGF signalling predates the origin of the bilaterians . Furthermore, it was recently suggested that posterior growth through terminal addition was already present in the bilaterian ancestor . Thus, we may ask whether these properties have played a decisive role in sending evolution down the path of evolving sequential segmentation.
In the current study we aim to answer these questions. For this we substantially extended a previously used in silico model . Rather than superimposing a particular growth pattern, we incorporate a gene controlling cell division and let evolution determine the type of tissue growth dynamics that arises. Then, by varying whether or not a stable posterior signalling centre is present in simulations, we can investigate the role of such a signalling centre on the type of growth and segmentation that evolves. We thus substantially expanded the degrees of freedom available to the evolutionary process, allowing us to investigate under which conditions sequential segmentation is the most likely evolutionary outcome.
We observe two predominant evolutionary outcomes: sequential segmentation with posterior growth and simultaneous segmentation involving tissue-wide bursts of divisions. We find that the likelihood with which the strategies evolve depends on the type of imposed morphogen dynamics and the strength and timing of an evolutionary pressure for determinate growth. We show that a self-organised posterior signal is more difficult to evolve than a developmental strategy which does not rely on such a posterior centre. From this we conclude that the prior evolution of a posterior signalling centre has played a decisive role in determining the evolution of sequential segmentation. Furthermore, we demonstrate that an evolutionary pressure for determinate growth reduces the likelihood of evolving sequential segmentation. When we apply this evolutionary pressure after sequential growth and segmentation have evolved, a mechanism to stop growing can evolve which is coordinated with the pre-existing sequential segmentation. We therefore propose that the order of evolutionary events is key to inferring the likelihood of particular developmental strategies. Reversing the argument, our work strongly suggests that a posterior signalling zone evolved prior to segmentation and that sequential growth and segmentation evolved prior to determinate growth.
Developmental tool kit and initial conditions There are 16 types of genes, indicated with a number from 0 to 15.
Gene 1 and gene 2 are signalling genes, responsible for direct, membrane bound cell–cell signalling (similar to e.g. Delta–Notch signalling). This direct cell–cell signalling is implemented as follows: if a gene has TFBS of type 1 or 2 in its upstream region, the expression of that gene in a particular cell is regulated by the levels of protein type 1 or 2 in its directly neighbouring cells, while its own intracellular levels of these proteins have no impact on the expression of that gene but only on that of genes in neighbouring cells. If cell–cell signalling is switched off in a simulation, genes 1 and 2 function as normal genes.
Gene 4 is the division gene when it is highly expressed (protein level >80. a.u.), the cell may divide with high probability (\(p=0.975\)). Upon division, the level of only the division gene is halved in the resulting two daughter cells (Fig. 1c).
Gene 5 is the segmentation gene, whose final pattern of gene expression along the body axis determines the fitness of the organism.
Individuals start their development with a short row of 10 cells, where the posterior cell forms the primordial “growth zone” in which the morphogen level is high; in the remaining nine cells (the “head” that is assumed to have evolved prior), the morphogen is absent (Fig. 1b). At the start of development genes 6 and 7 are uniformly expressed in the zygote, while other genes have an initial expression level of 0. Throughout development, the protein levels are updated according to the network (Eq. 1).
Fitness evaluation At the end of development (after a fixed number of time steps), the number of well-formed segments determines an individual’s fitness. A segment is defined by the average expression pattern of the segmentation gene over the last 20 or 40 developmental steps (Fig. 1d). This averaging helps ensure the evolution of temporally stable segmental patterning, as it will not reward oscillatory segmentation that fails to converge on a constant spatial pattern (as occurred in ). Segments should be at least seven cells wide, and boundaries between segments should consist of a clear transition of the expression of the segmentation gene from a high to a low level, or vice versa, within five cells (similar to earlier definitions [24, 26]. The number of too narrow segments is subtracted from the number of well-formed ones, reducing the fitness. To further ensure stability of the final developmental pattern, we apply an additional fitness penalty for the amount of variance of the pattern from the average (pattern instability) within the final 20 developmental steps.
30 × 30
Evolutionary time steps
Initial # agents
Developmental time steps
The number of integration steps
Integration step size
Fourth-order Runge–Kutta integration
Morphogen decay rate
Only for persistent posterior morphogen of which 9 form the head
Initial tissue size
Gene and protein dynamics
Gene product decay rate
Hill constant of the TFBS
Nr of gene types
Note that with the gene, also its TFBS is duplicated
TFBS weight change
TFBS type change
Spontaneous emergence of new TFBS
G: penalty per gene
T: penalty per TFBS
Bonus for final tissue size
0 or 0.1
Per cell added by division
Penalty for exceeding target size
For each cell more than target size
Period over which gene expression stability and sometimes number of late-stage divisions is measured
U: expression variance penalty
Penalty per cell that has a variance in segmentation gene level >5.0 during the control period
Initial conditions, mutations and simulations The population is initialised with 50 genetically identical individuals. The population resides in a grid of size 30 × 30, imposing a maximum to the population size of 900 individuals. The genome of the initial individuals contains a single copy of each gene, in randomised order and with an average of 2 TFBS of random type upstream. Individuals compete for reproduction into a neighbouring empty spot. Those with a higher fitness have a larger probability of being selected. Specifically, an individual’s chance to reproduce is proportional to its fitness divided by the sum over the fitness of itself and the other individuals neighbouring the empty position. Death occurs with a constant probability. Upon reproduction, the genome is mutated via duplications and deletions of both genes and TFBS (Fig. 1d). TFBS may also mutate their type (which protein binds), weight (activating or repressing) and Hill constant, and new TFBS may appear de novo as an innovation. Genes may mutate their maximum activation level E and decay rate d. Gene duplication also copies the associated TFBS and results in multiple genes of the same type. The expression of all genes of the same type therefore contributes to the expression level of a single protein. Note that since there are no mutations that change the gene type, gene duplication cannot be followed by subsequent divergence in our model.
For each set of model settings and parameter values, we run 50 simulations. Each simulation yields one particular growth and segmentation strategy with only minor variations within the population. Therefore, we only assess one fit individual per simulation. We consider a simulation successful when the fittest individuals at the end of the simulation can generate more than three segments.
Pruning Because an evolved genome consists partly of redundant interactions, we prune the genomes via a repeated process of removing genes and binding sites in the genome, while keeping the tissue size and final spatial expression pattern of the segmentation gene the same . This makes it easier to analyse the network by eye and identify the role of the special genes such as the division gene and the segmentation gene. All networks depicted are pruned.
Assessment of phenotypic variability We have only one source of noise in our simulations, namely the small probability that a cell with high division gene expression does not immediately divide. Still, this noise may cause some variability in the number of segments that are formed, even if the genome remains exactly the same. We call this phenotypic variability and assess this by repeating the development of an evolved individual 50 times and counting the number of good and malformed segments formed each time. We display the results of this repeated development in a histogram, where we indicate the number of malformed segments with lighter bars. When an individual evolved a mechanism that makes the same number of good segments in 40 or more times out of 50 repeated developments, we call it “developmentally robust”. Note that this differs from the probability of breeding true (mutational robustness): we do not assess how mutations influence the likelihood of producing the same number of segments, as done in .
Evolved developmental strategies
Transient posterior morphogen
No CCS, no noise
2 (Fig. 4)
With CCS, no noise
1 (Additional file 4: Figure S4)
Persistent posterior morphogen
Tissue size selection
6 (Fig. 7)
31 (Fig. 7)
Division gene dynamics In some figures, we display the dynamics of the division gene in the posterior-most cell (containing a high morphogen level). These dynamics result from regulation of the division gene by other genes and from halving of the division gene upon division. In order to see the effect of regulation more clearly, we also show the network dynamics when the halving of the division gene is left out.
Body axis segmentation is, like most developmental patterning, a complex phenomenon involving processes ranging from the subcellular to the organism level scale. In this study we simulate the evolution of body axis segmentation and therefore need to simulate development in a population of individuals across many generations. To keep our model tractable both in terms of simulation time and for analysing the evolutionary trajectories of the developmental processes, we substantially simplify the developmental processes in our model relative to real-world developmental processes, but incorporate those properties we deem necessary for studying the evolution of segmentation. Below we detail the three major simplifications, their potential consequences, and why we think these simplifications are justified.
The most obvious simplification is the 1D nature of the model, so that cell divisions automatically lead to an elongated body axis regardless of where they occur in the tissue basically assuming that the axiality is already defined. In reality, developmental patterning occurs in a 2D or 3D tissue, where complex symmetry-breaking events during early development are essential for setting up the anterior–posterior and dorsal–ventral axis. Furthermore, animal axis elongation often involves cellular motility and adhesion properties that are not included in this model. This limits the self-organisation potential of the developmental processes evolving in our model. However, since symmetry breaking is an ancestral property preceding bilaterian evolution , we can safely assume that it already existed before segmentation evolved.
We model a cellularised environment in which morphogen gradients are set up through decay dynamics, and signalling is limited to direct receptor–ligand type cell–cell signalling. Neither the morphogen nor any of the other gene products are subject to diffusion. Diffusion plays an important role in Turing-type patterning  and in setting up the morphogen gradients dictating early Drosophila segmentation . Still, in Drosophila genes downstream of the morphogen gradients do not require diffusion, and so far no Turing patterns have been found to underlie animal segmentation. Segmentation usually takes place in a cellularised environment, in which the role of diffusion is necessarily restricted to short distances or combined with other gradient establishing mechanisms such as slow mRNA and protein decay . Thus, we are confident that we do not exclude any major real-world segmentation mechanisms from evolving in our model and that leaving out diffusion is justified.
Finally, our model only incorporates gene expression regulation through combinatorial TF regulation on a single regulatory region, ignoring several other factors that may influence gene expression. In vertebrates for instance, the timing of Hox gene mediated axial patterning, and its coordination with segmentation, involves chromosome looping, epigenetic histone and DNA modifications, and cluster level gene regulation . Still, the goal of our study is to explore the evolution of gene expression dynamics, rather than to mimic how these dynamics are precisely regulated. The transcription factor-based regulatory network has sufficient degrees of freedom to allow the evolution of the diverse gene expression dynamics (such as oscillations) that underlie real-world segmentation processes, while supporting computational efficiency and analytical tractability.
Evolutionary strategies with transient posterior signal
To investigate the relevance of the prior existence of a stable posterior signalling centre and the morphogen gradients emanating from it for the evolution of posterior growth and sequential segmentation, we performed simulations that do and simulations that do not superimpose the existence of such a signalling centre. We start with an exploration of evolving segmentation strategies in the absence of a superimposed morphogen gradient (Fig. 3a, inset). Instead, we assume transient expression of gene 0 (the “morphogen”), restricted to the posterior-most cell of the embryo, and subject to decay in all cells. As a consequence, this gene will have the same level in the posterior cell as in all cells that descended from it and information on tissue polarity becomes quickly diluted. Under these conditions, a stable posterior signalling centre would have to evolve from scratch by evolving regulation of this transient signal to generate a stable posterior morphogen gradient (rather than being automatically present [24, 26]). Alternatively, a segmentation mechanism could evolve which does not rely on a persistent posterior signal.
We perform four sets of 50 simulations: with/without cell–cell signalling (CCS) and with/without division noise (first four rows in Table 2). The simulations without CCS and without noise form a negative control group which does not have any symmetry-breaking mechanism: indeed, segmentation never evolves. In the remaining sets with either noise, CCS or both, segmentation does evolve. The vast majority of successful simulations (yielding more than three segments) evolves a segmentation strategy in which the tissue grows via one or more short-lived tissue-wide burst of divisions (Fig. 3a). The segments all appear roughly at the same time after the burst of divisions; we call this simultaneous segmentation.
In the absence of CCS and presence of division noise, this simultaneous segmentation mechanism typically yields high phenotypic variability, often generating few segments and only occasionally producing many segments (e.g. Fig. 3b, see “Methods”). The segments are often irregular in size, with some much wider than others. The segmentation mechanism uses the stochastic delay of division in a few cells early in development, which changes the dynamics in those cells sufficiently to differentiate them from their neighbours (Fig. 3c, d). This mechanism therefore does not resemble Drosophila-type simultaneous segmentation but rather reflects the fact that the evolutionary process is free to evolve any possible growth and patterning modes.
Altogether, without a superimposed posterior morphogen gradient we obtain a nearly 100 % bias towards simultaneous growth and patterning, with the rare appearance of polarised growth. We therefore next test whether a polarised growth dynamics evolves more frequently if we select for tissue size but not for segmentation, thereby reducing the complexity of the selection target.
Our results show that without a superimposed posterior morphogen gradient, evolution is unlikely to result in posterior growth and sequential segmentation.
Evolutionary strategies with persistent posterior signal
Next, we performed two sets of simulations with a persistent posterior signal, in the form of a superimposed posterior morphogen gradient: one set with and one without cell–cell signalling, and both with division noise (Table 2). To achieve this, the posterior-most cell now receives a morphogen that is subject to decay in all cells except this posterior-most cell. In these simulations, we find two qualitatively different strategies. The majority of simulations (53 out of 100) evolves a posterior growth zone combined with sequential segmentation, while the tissue-wide burst with simultaneous segmentation observed in the previous section is now less common (30 out of 100). In the simulations with CCS we also find combinations of the simultaneous and sequential mechanism, where the first few segments are generated using a simultaneous mechanism and the remaining segments arise through posterior growth combined with gene expression oscillations (5 simulations).
In the absence of CCS, a posterior morphogen gradient improves the developmental robustness of the evolved simultaneous patterning mechanisms (examples in Additional file 3: Figure S3; see Table 2). Cells may use differences in morphogen concentration rather than the differences arising through stochastic cell division for segmental patterning (in fact, now division noise is the source of phenotypic variability, not the patterning mechanism). While adding CCS to simulations with transient posterior morphogen decreased phenotypic variability, in the presence of persistent posterior morphogen no further improvement was observed upon addition of CCS.
So far, we selected for increasing numbers of segments, thus only implicitly selecting for tissue growth. However, it can be hypothesised that body axis elongation—even in the absence of subdivision into segments—already confers an evolutionary advantage. We therefore compare the previously described simulations in which we selected only for segmentation (the set without CCS), to simulations in which we independently select for both axial elongation (to a particular target size) and segmentation. While the number of simulations yielding sequential mechanisms is the same (31), we now find that eight simulations yield the combined simultaneous and sequential strategy, and only six simulations yield fully simultaneous segmentation. Thus, the bias towards sequential segmentation has become somewhat stronger.
We conclude that, of the possible evolutionary options to segment a tissue, posterior growth coupled to sequential segmentation has a higher potential to be a robust patterning mechanism and is capable of generating more and more regularly shaped segments. In addition, when incorporating an ancestral posterior signalling centre involved in body axis polarity, it is also the most likely evolutionary outcome. This likelihood slightly increases when body axis extension evolves prior to segmentation.
Evolving determinate growth
From the previous section, it is clear that posterior growth with sequential segmentation is the most successful of the possible developmental strategies: it evolves more often, it is able to form many and regularly shaped segments, and it has the potential to be developmentally robust. So far, we did not take into account that the evolved sequential segmentation mechanisms do not terminate growth at the end of development. Instead they evolve a growth rate that is tuned to allow them to grow to the target size within the constant, superimposed duration of development. If this duration of development were to be extended, larger individuals with a larger number of segments would automatically arise. While there are indeed bilaterian animals (like many annelids) which do continue growing indefinitely , most animals stop growing and making segments, for instance vertebrates and insects have a determinate number of segments and roughly determinate growth. We therefore decided to include selection for determinate growth, by applying a fitness penalty for division during the last 20 time steps (no CCS). (Note that the definition applied here for determinate and indeterminate growth is somewhat different from definitions used elsewhere . See also the Fig. 2).
We reason that by applying the selection for determinate growth from the start of the simulation, we implicitly select for simultaneous growth, which is indeed determinate. To prevent this bias, sequential segmentation would have to have evolved before the appearance of this selection pressure. It also seems biologically reasonable to assume that determinate growth is a secondary trait: when comparing segmented animals with indeterminate and determinate growth, it seems that at least in arthropods the clades with determinate growth have more evolutionary derived, complex body plans. We speculate that determinate growth becomes more important upon evolution of segment specification, where e.g. locomotive appendages are limited to trunk segments and the abdomen is unsupported. To test the idea of secondary selection for determinate growth, we extract individuals from 20 earlier simulations in which sequential segmentation evolved without the pressure to stop growing; then we continue their evolution in the presence of this pressure. The outcome of these continued simulations depends on the strength of the evolutionary pressure to stop growing. If the pressure is too low, determinate growth does not evolve often (Fig. 8b). If instead the pressure is too high, the potential for growth and segmentation is often transiently lost after which a simultaneous mechanism evolves instead; we do not observe smooth transitions from sequential to simultaneous segmentation (Fig. 8b). However, between these two extremes lies a parameter region in which one-third of simulations evolve the capacity to stop growing while maintaining posterior growth and sequential segmentation (Fig. 8b). Thus, our hypothesis is confirmed, delaying selection for determinate tissue growth to a later evolutionary stage does indeed more often yield the evolution of sequential growth and segmentation combined with determinate growth.
Developmental programmes incorporating the first stopping mechanism (gene expression build-up) become slightly less robust compared to the original sequential segmentation mechanism without stopping growth, as stochastic divisions may influence the time at which the growth stopping protein level is being reached (Fig. 9c). Developmental programmes applying the second mechanism (stochastic threshold passing) become significantly less robust, which logically follows from the fact that they rely on the stochasticity of divisions to determine when to stop (Fig. 9c). Note that in both cases segment size does remain regular.
We found one main alternative developmental strategy besides sequential segmentation: simultaneous segmentation, in which after a short tissue-wide burst of divisions all segments appears roughly at the same time. This simultaneous mechanism is not similar to Drosophila-like segmentation, where a hierarchy of gene regulation robustly creates regular-sized segments. Note that the Drosophila strategy likely evolved secondarily, from an initial sequential segmentation mode, so one should neither aim nor expect a Drosophila-like segmentation to evolve from scratch in our simulations. Rather, the evolution of a quite different type of simultaneous segmentation in our simulations is a result of the freedom of the evolutionary process, which we use to distinguish evolutionary scenarios. The simultaneous strategy that evolved in our simulations generates irregularly sized segments, and the number of segments in genotypically identical individuals tends to be variable. In contrast, the evolved sequential segmentation generates a large number of regularly sized segments in a robust, reproducible manner, thus leading to larger fitness values at the end of evolutionary simulations. A number of subtypes of posterior growth and sequential segmentation evolved; the most notable mechanism involves regular, segmental oscillation-dependent bursts of cell division in which two segments are down simultaneously—another benefit of the larger degrees of freedom of the model.
Stable posterior signalling is a prerequisite for sequential segmentation We showed that evolution of terminal addition type posterior growth is highly unlikely in the absence of persistent posterior signalling, independent of whether we selected for segmentation or body axis elongation. Under these conditions, the potential for symmetry breaking is restricted to the early phases of development, generating a bias in favour of an early tissue-wide burst of divisions and against posterior growth. In the absence of cell–cell signalling, simulations relied on the stochasticity of divisions to generate segments. In the presence of cell–cell signalling, lateral inhibition type patterning is used to pattern segments during the tissue-wide division burst. Our results thus suggest that the evolution of a posterior signalling centre is a crucial prerequisite for the evolution of posterior growth and sequential segmentation. Given the presence of a posterior signalling centre in all bilaterians as well as cnidarians, it can be safely assumed to represent an ancestral property [27, 37, 38]. Thus, we can reformulate our findings and state instead that the prior evolution of a posterior signalling centre provided a strong bias towards the evolution of posterior growth and sequential segmentation.
Determinate growth as a secondary trait In simulations incorporating a persistent posterior morphogen signal, selection for determinate growth completely reversed the evolutionary bias from sequential to simultaneous segmentation. We found that to evolve posterior yet determinate growth, the selection for determinate growth had to occur secondarily, after the evolution of posterior growth and sequential segmentation. Simple segmented animals such as millipedes and annelids contain large numbers of highly similar segments, and many annelids appear to keep adding segments throughout their life . In contrast, insects and vertebrates develop a smaller, constant number of highly specialised segments after which posterior growth is terminated. We thus hypothesise that a constant segment number evolved secondarily and was only selected for once segment specialisation arose and locomotive capabilities became restricted to a limited number of segments. Consistent with this, HOX genes, which are crucial in segment specialisation, appear to be involved in terminating posterior growth . As an intermediate form, myriapods and the extinct trilobites stop adding segments when reaching maturity, but the final number of segments is variable . This has been linked to their limited segment specialisation, where the exact number of segments is not that important. This is reminiscent of the form of determinate growth that evolves in our simulations, which is not robust and yields variable segment numbers.
Still, how to explain the fact that many unsegmented and metameric animals display determinate growth, for instance C. elegans? Assuming an unsegmented bilaterian ancestor, determinate growth may have evolved prior to sequential segmentation. In the current study, we did not explicitly test for this; however, we expect that even if we selected for only axial tissue growth without segmentation, simultaneous growth would arise if we also immediately selected for determinate growth. Given that ancestral state reconstructions suggest that terminal addition is an ancestral bilaterian trait [9, 14] while the evidence is less conclusive for sequential segmentation, we would expect that determinate growth at least evolved secondary to posterior growth. Alternatively, the presence of determinate growth in unsegmented and metameric organisms can be explained by the presence of a bilaterian ancestor displaying terminal addition, sequential segmentation and determinate growth with many lineages subsequently completely or partly losing segmentation. Finally, a less parsimonious scenario involves an unsegmented, indeterminately growing bilaterian ancestor, with parallel evolution of either determinate growth alone, or following sequential segmentation in several lineages.
In our current model we observed two mechanisms to stop posterior growth: one depending on stochastic changes in an oscillating gene, making it very non-robust, and the other depending on the gradual build-up of a slowly decaying gene, yielding lower phenotypic variability. This latter strategy resembles a hypothetical mechanism proposed by Meinhardt for the sequential activation of HOX genes [30, 40]. Considering the origin of the HOX cluster from tandem duplication of an early HOX gene , it can be envisioned that an early gene involved in regulating growth became transformed through duplication into a sequentially activated HOX cluster in which gene order (i.e. posterior-most HOX gene active) rather than level of a single gene now can act as a robust growth termination criterion. An interesting subject for future studies would thus be to investigate whether under combined selection for both segmentation and HOX-like specialisation domains, a HOX-type control of growth evolves. Such evolutionary outcomes may provide important answers in the puzzle of how the complex hierarchical network of Drosophila evolved from a sequentially segmenting predecessor; given the relatedness of gap and HOX genes and the suggested ancestral role for gap genes in growth control.
In summary, we proposed an order of evolutionary events and selection pressures involved in generating posterior growth, sequential segmentation and determinate growth. First, we provided evidence that the prior evolution of a stable posterior signalling centre has played a decisive role in evolving terminal addition and sequential segmentation. Then, we showed that the evolution of sequential segmentation combined with determinate growth can only take place by adding the selection pressure for determinate growth secondarily. Our study demonstrates that varying the onset of selection pressures can be a powerful tool in investigating the likely order of evolutionary events.
transcription factor binding site
RMAV and KHWJT designed the study, RMAV performed the simulations, and RMAV, PH and KHWJT analysed the results and wrote the manuscript. All authors read and approved the manuscript.
We thank Enrico Sandro Colizzi, Hilje Doekes, Bram van Dijk, Rutger Hermsen and Thomas Cuypers for helpful suggestions and discussion of the manuscript.
The authors declare that they have no competing interests.
RMAV was supported by the “Focus en Massa” program from Utrecht University. The funding body had no role in the design of the study, collection and analysis of the data, or writing of the manuscript.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
- Davis GK, Patel NH. The origin and evolution of segmentation. Trends Genet. 1999;15(12):68–72. doi:10.1016/S0168-9525(99)01875-2.View ArticleGoogle Scholar
- Balavoine G, Adoutte A. The segmented urbilateria: a testable scenario. Integr Comp Biol. 2003;43(1):137–47. doi:10.1093/icb/43.1.137. http://icb.oxfordjournals.org/content/43/1/137.full.pdf+html.
- Peel A, Akam M. Evolution of segmentation: rolling back the clock. Curr Biol. 2003;13(18):708–10. doi:10.1016/j.cub.2003.08.045.View ArticleGoogle Scholar
- Couso JP. Segmentation, metamerism and the Cambrian explosion. Int J Dev Biol. 2009;53:1305–16. doi:10.1387/ijdb.072425jc.View ArticlePubMedGoogle Scholar
- Budd GE. Why are arthropods segmented? Evol Dev. 2001;3(5):332–42. doi:10.1046/j.1525-142X.2001.01041.x.View ArticlePubMedGoogle Scholar
- Seaver EC. Segmentation: mono- or polyphyletic? Int J Dev Biol. 2003;47:583–95.PubMedGoogle Scholar
- Minelli A, Fusco G. Evo-devo perspectives on segmentation: model organisms, and beyond. Trends Ecol Evol. 2004;19(8):423–9. doi:10.1016/j.tree.2004.06.007.View ArticlePubMedGoogle Scholar
- Tautz D. Segmentation. Dev Cell. 2004;7(3):301–12. doi:10.1016/j.devcel.2004.08.008.View ArticlePubMedGoogle Scholar
- Jacobs DK, Hughes NC, Fitz-Gibbon ST, Winchell CJ. Terminal addition, the cambrian radiation and the phanerozoic evolution of bilaterian form. Evol Dev. 2005;7(6):498–514. doi:10.1111/j.1525-142X.2005.05055.x.View ArticlePubMedGoogle Scholar
- Blair SS. Segmentation in animals. Curr Biol. 2008;18(21):991–5. doi:10.1016/j.cub.2008.08.029.View ArticleGoogle Scholar
- Wanninger A, Kristof A, Brinkmann N. Sipunculans and segmentation. Commun Integr Biol. 2009;2(1):56–9. doi:10.4161/cib.2.1.7505.View ArticlePubMedPubMed CentralGoogle Scholar
- Chipman AD. Parallel evolution of segmentation by co-option of ancestral gene regulatory networks. BioEssays. 2010;32(1):60–70. doi:10.1002/bies.200900130.View ArticlePubMedGoogle Scholar
- Richmond DL, Oates AC. The segmentation clock: inherited trait or universal design principle? Curr Opin Genet Dev. 2012;22(6):600–6. doi:10.1016/j.gde.2012.10.003 (Genetics of system biology).View ArticlePubMedGoogle Scholar
- Gold DA, Runnegar B, Gehling JG, Jacobs DK. Ancestral state reconstruction of ontogeny supports a bilaterian affinity for Dickinsonia. Evol Dev. 2015;17(6):315–24. doi:10.1111/ede.12168.View ArticlePubMedGoogle Scholar
- Rivera A, Weisblat D. And Lophotrochozoa makes three: Notch/Hes signaling in annelid segmentation. Dev Genes Evol. 2009;219(1):37–43. doi:10.1007/s00427-008-0264-6.View ArticlePubMedGoogle Scholar
- Williams T, Blachuta B, Hegna TA, Nagy LM. Decoupling elongation and segmentation: Notch involvement in anostracan crustacean segmentation. Evol Dev. 2012;14(4):372–82. doi:10.1111/j.1525-142X.2012.00555.x.View ArticlePubMedGoogle Scholar
- Shankland M, Seaver EC. Evolution of the bilaterian body plan: what have we learned from annelids? Proc Natl Acad Sci. 2000;97(9):4434–7. doi:10.1073/pnas.97.9.4434. http://www.pnas.org/content/97/9/4434.full.pdf.
- Balavoine G. Segment formation in annelids: patterns, processes and evolution. Int J Dev Biol. 2014;58:469–83. doi:10.1387/ijdb.140148gb.View ArticlePubMedGoogle Scholar
- Peel A. The evolution of arthropod segmentation mechanisms. BioEssays. 2004;26(10):1108–16. doi:10.1002/bies.20097.View ArticlePubMedGoogle Scholar
- Bénazéraf B, Pourquié O. Formation and segmentation of the vertebrate body axis. Annu Rev Cell Dev Biol. 2013;29(1):1–26. doi:10.1146/annurev-cellbio-101011-155703.View ArticlePubMedGoogle Scholar
- Mayer G, Kato C, Quast B, Chisholm RH, Landman KA, Quinn LM. Growth patterns in onychophora (velvet worms): lack of a localised posterior proliferation zone. BMC Evol Biol. 2010;10(339):1–12. doi:10.1186/1471-2148-10-339.Google Scholar
- Nakamoto A, Hester SD, Constantinou SJ, Blaine WG, Tewksbury AB, Matei MT, Nagy LM, Williams TA. Changing cell behaviours during beetle embryogenesis correlates with slowing of segmentation. Nat Commun. 2015;6(6635):1–9. doi:10.1038/ncomms7635.Google Scholar
- Graham A, Butts T, Lumsden A, Kiecker C. What can vertebrates tell us about segmentation? EvoDevo. 2014;. doi:10.1186/2041-9139-5-24.PubMedPubMed CentralGoogle Scholar
- François P, Hakim V, Siggia ED. Deriving structure from evolution: metazoan segmentation. Mol Syst Biol. 2007. doi:10.1038/msb4100192.PubMedPubMed CentralGoogle Scholar
- Fujimoto K, Ishihara S, Kaneko K. Network evolution of body plans. PLoS One. 2008;3(7):2772. doi:10.1371/journal.pone.0002772.View ArticleGoogle Scholar
- ten Tusscher KH, Hogeweg P. Evolution of networks for body plan patterning; interplay of modularity, robustness and evolvability. PLoS Comput Biol. 2011;7(10):1002208. doi:10.1371/journal.pcbi.1002208.View ArticleGoogle Scholar
- Niehrs C. On growth and form: a cartesian coordinate system of wnt and bmp signaling specifies bilaterian body axes. Development. 2010;137(6):845–57. doi:10.1242/dev.039651. http://dev.biologists.org/content/137/6/845.full.pdf.
- Crombach A, Hogeweg P. Evolution of evolvability in gene regulatory networks. PLoS Comput Biol. 2008;4(7):1–13. doi:10.1371/journal.pcbi.1000112.View ArticleGoogle Scholar
- Salazar-Ciudad I, Newman SA, Solé RV. Phenotypic and dynamical transitions in model genetic networks I. Emergence of patterns and genotype–phenotype relationships. Evol Dev. 2001;3(2):84–94. doi:10.1046/j.1525-142x.2001.003002084.x.View ArticlePubMedGoogle Scholar
- Meinhardt, H.: Models of biological pattern formation. London: Academic Press; 1982. http://www.eb.tuebingen.mpg.de/research/emeriti/hans-meinhardt/82-book/bur82.html
- Pankratz MJ, Jäckle H. Making stripes in the drosophila embryo. Trends Genet. 1990;6:287–92. doi:10.1016/0168-9525(90)90234-W.View ArticlePubMedGoogle Scholar
- Aulehla A, Pourquié O. Signaling gradients during paraxial mesoderm development. Cold Spring Harbor Perspect Biol. 2010;2(2):1–17. doi:10.1101/cshperspect.a000869. http://cshperspectives.cshlp.org/content/2/2/a000869.full.pdf+html.
- Mallo M, Alonso CR. The regulation of hox gene expression during animal development. Development. 2013;140(19):3951–63. doi:10.1242/dev.068346. http://dev.biologists.org/content/140/19/3951.full.pdf.
- Crampin EJ, Hackborn WW, Maini PK. Pattern formation in reaction–diffusion models with nonuniform domain growth. Bull Math Biol. 2002;64(4):747–69. doi:10.1006/bulm.2002.0295.View ArticlePubMedGoogle Scholar
- Harper JL, Rosen BR, White J. Growth and form of modular organisms. Philos Trans R Soc Lond. 1986;313(1159):1–250.View ArticleGoogle Scholar
- ten Tusscher KHWJ. Mechanisms and constraints shaping the evolution of body plan segmentation. Eur Phys J E. 2013;36(5):1–12. doi:10.1140/epje/i2013-13054-7.Google Scholar
- Petersen CP, Reddien PW. Wnt signaling and the polarity of the primary body axis. Cell. 2009;139(6):1056–68. doi:10.1016/j.cell.2009.11.035.View ArticlePubMedGoogle Scholar
- Martin BL, Kimelman D. Wnt signaling and the evolution of embryonic posterior development. Curr Biol. 2009;19(5):215–9. doi:10.1016/j.cub.2009.01.052.View ArticleGoogle Scholar
- Young T, Rowland JE, van de Ven C, Bialecka M, Novoa A, Carapuco M, van Nes J, de Graaff W, Duluc I, Freund J-N, Beck F, Mallo M, Deschamps J. Cdx and hox genes differentially regulate posterior axial growth in mammalian embryos. Dev Cell. 2009;17(4):516–26. doi:10.1016/j.devcel.2009.08.010.View ArticlePubMedGoogle Scholar
- Meinhardt H. Models for patterning primary embryonic body axes: the role of space and time. Semin Cell Dev Biol. 2015;42:103–17. doi:10.1016/j.semcdb.2015.06.005 (Claudins and time, space and the vertebrate body axis).View ArticlePubMedGoogle Scholar
- Kappen C, Schughart K, Ruddle FH. Two steps in the evolution of Antennapedia-class vertebrate homeobox genes. Proc Natl Acad Sci USA. 1989;86:5459–63.View ArticlePubMedPubMed CentralGoogle Scholar
- Sarrazin AF, Peel AD, Averof M. A segmentation clock with two-segment periodicity in insects. Science. 2012;336(6079):338–41. doi:10.1126/science.1218256. http://www.sciencemag.org/content/336/6079/338.full.pdf.